Copyright 2011 by Oriana Bandiera, Luigi Guiso, Andrea Prat, and Raffaella Sadun

Working papers are in draft form. This working paper is distributed for purposes of comment and
discussion only. It may not be reproduced without permission of the copyright holder. Copies of working
papers are available from the author.


What Do CEOs Do?

Oriana Bandiera
Luigi Guiso
Andrea Prat
Raffaella Sadun




Working Paper

11-081

What Do CEOs Do?

Oriana Bandiera
London School of Economics
Luigi Guiso
European University Institute and EIEF
Andrea Prat
London School of Economics
Raaella Sadun
Harvard University
February 25, 2011
Abstract
We develop a methodology to collect and analyze data on CEOs time use. The idea
sketched out in a simple theoretical set-up is that CEO time is a scarce resource and its
allocation can help us identify the rms priorities as well as the presence of governance
issues. We follow 94 CEOs of top-600 Italian rms over a pre-specied week and record the
time devoted each day to dierent work activities. We focus on the distinction between
time spent with insiders (employees of the rm) and outsiders (people not employed by
the rm). Individual CEOs dier systematically in how much time they spend at work
and in how much time they devote to insiders vs. outsiders. We analyze the correlation
between time use, managerial eort, quality of governance and rm performance, and
interpret the empirical ndings within two versions of our model, one with eective and
one with imperfect corporate governance. The patterns we observe are consistent with
the hypothesis that time spent with outsiders is on average less benecial to the rm and
more benecial to the CEO and that the CEO spends more time with outsiders when
governance is poor.

We thank seminar participants at Catholic University of Milan, Columbia University, EEA Meetings,
Harvard Business School, Imperial College, IMT Lucca, LSE, MIT, New York University, NIESR, Universitat
Pompeu Fabra, Stockholm (IIES), and the University of Sussex for useful comments. We are grateful to
Tito Boeri, Daniel Ferreira, Luis Garicano, Steve Pischke, Imran Rasul, and Fabiano Schivardi for useful
suggestions. This research was made possible by generous funding from Fondazione Rodolfo Debenedetti.
1
1 Introduction
Time management has been at the core of the research on management for almost fty years.
Peter Drucker (1966) makes it the centerpiece of his celebrated book on eective management:
Eective executives know that time is the limiting factor. The output of any process is set
by the scarcest resource. In the process we call accomplishment, this is time.
More recently, the importance of time constraints for people at the top of organiza-
tional hierarchies has been recognized by a number of economic models, like Radner and Van
Zandts work on hierarchies (surveyed in Van Zandt, 1998), Bolton and Dewatripont (1994),
and Garicano (2000). Intellectual tasks information processing in Radner-Van Zandt, com-
munication in Bolton-Dewatripont, problem solving in Garicano require managerial time,
which is particularly scarce at the top of the organization.
As a consequence, a key question in organization economics is whether and how managers
allocate their time to maximize rm performance. The theoretical and practical interest,
however, are not matched by adequate empirical evidence, as rm datasets typically contain
no information on managerial time use, while existing studies of managerial time use are
mostly based on very few observations and have not been matched to rm level outcomes.
This paper makes a step towards lling this gap by developing a methodology to system-
atically collect information on how CEOs allocate their time between dierent work activities.
We develop a time-use diary for CEOs and we use it to record the time allocation of a sample
of 94 CEOs belonging to leading Italian companies in various industries over a pre-selected
work week. A minimalistic model of managerial use of time guides our interpretation of ob-
served dierences in time allocation patterns, and how these are related to the rms priorities
and to dierences between the rms and the CEOs interests.
To collect the time use data, we ask the CEOs personal assistant (PA) to keep a diary of
the activities performed by the CEO during a pre-specied week (from Monday to Friday).
The PA has full information on the CEOs schedule and he/she has a good understanding of
the functions of the people that his/her boss interacts with.
1
The PA reports overall working
time and records in a diary all the activities of the CEO that last longer that 15 minutes
detailing their characteristics, in particular information on other participants.
To operationalize the concept of time allocation, we group activities according to whether
they involve employees of the rm (insiders) or only people external to the rm (outsiders).
Insiders are listed by functional area, for instance nance, marketing, or human resources.
1
Roxanne Sadowski, Jack Welchs long-term PA, makes this abundantly clear: For more than fourteen
years, Ive been human answering machine, auto dialer, word processor, ltering system, and fact checker;
been a sounding board, schlepper, buddy, and bearer of good and bad tidings; served as a scold, diplomat,
repairperson, cheerleader and naysayer; and performed dozens of other roles under the title of assistant.
2
Outsiders are grouped in categories CEOs normally interact with, for instance suppliers,
investors, or consultants. The insider/outsider classication is ideal for our purposes for two
reasons. First, it captures the essence of the CEOs job: The CEO is the link between the
Inside that is the organization, and the Outside of society, economy, technology, markets, and
customers (Drucker, cited in Laey 2009). Second, it is a dimension over which the CEOs
and the rms interests might be misaligned. As a matter of fact, the optimal inside/outside
balance is the subject of controversy in the management, nance and economics literature. At
one end of the spectrum, a widely held view is that since the CEO is the "public face" of the
company, spending time outside the rm is an important (if perhaps not the most important)
role of the CEO.
2
At the other end, however, an increasingly popular view points out that
time spent with outsiders might mostly benet the CEO without contributing to creating
value for the rm. Khurana (2002), for instance, argues that CEOs have strong incentives
to seek visibility, by cultivating personal connections with inuential business leaders. In
line with this, Malmendier and Tate (2009) show that when their power vis-a-vis the rm
increases, CEOs spend more time outside the rm in activities, such as writing books and
playing golf, and that this shift in activities does not contribute to rms performance.
3
Our data reveal that, as expected, CEOs spend the majority of their time with other
people (85%). Of these most are employees of the same rm, but many are not. On average,
CEOs spend 42% of their time with insiders only, 25% with both insiders and outsiders and
16% with outsiders alone. More interestingly, these averages hide a great deal of heterogeneity,
both in terms of number of hours worked and time allocation. A majority of CEOs spend
very little time (less than 5 hours per week) alone with outsiders, but over 10% of them
spend over 10 hours a week. Our empirical analysis is aimed at making sense of the wide
heterogeneity on this dimension of time allocation choices.
Since the implication of dierent time allocation choices - and in particular the dier-
ential impact of time spent with insiders vs. outsiders - is a priori ambiguous, we develop
a minimalistic model of managerial time allocation that allows us to distinguish productive
activities from activities that mostly confer private benets to the CEO. We use this frame-
work to shed light on whether time with insiders and outsiders can be interpreted along these
lines. In our model, the CEO chooses how much time to devote to a number of possible
2
Procter & Gambles CEO A. G. Laey (2009) views the link with the outside as the raison dtre of the
top executive: The CEO alone experiences the meaningful outside at an enterprise level and is responsible
for understanding it, interpreting it, advocating for it, and presenting it so that the company can respond in
a way that enables sustainable sales, prot, and total shareholder return (TSR) growth.
3
Khurana (2002)s view is that the market for CEOs is deeply awed. The search for external, charismatic
corporate saviors leads boards and the executive search companies they employ to give their preference
to highly visible personalities that can more easily be "sold" to the rm shareholders. In turn, this creates an
incentive for CEOs to seek visibility, by cultivating personal connections with inuential business leaders.
3
work-related activities, or to leisure. Each of the work activities yield non-negative benets
to the rm and to the CEO. For instance, networking with clients might increase the rms
sale but also increase the chance that the CEO is oered a better job in the future. Suppose
activities can be roughly divided into the ones that benet mostly the rm and those that
benet mostly the CEO. In this set-up, the interpretation of dierences in time allocation
between these two sets of activities depends on the extent to which the CEOs and the rms
interests are aligned. When these are perfectly aligned, time allocation is entirely determined
by the rms objectives. In this case, the observed variation in time allocation is entirely de-
termined by optimal responses to dierent conditions faced by dierent rms. However, when
the CEOs and the rms interests are not perfectly aligned, time allocation is determined by
both the rms objectives and the CEOs objectives. In this case, the observed variation in
time allocation partially reects dierences in the quality of governance that determines the
alignment of the CEOs interests with the rms.
If the observed variation reects dierences in governance, the model makes precise that
the following three sets of correlations must hold:
1. CEOs who work longer hours, devote more time to activities that mostly benet the
rm and less time to activities that mostly yield private benets.
2. CEOs who work for rms with stronger governance devote more time to activities that
mostly benet the rm and less time to activities that mostly yield private benets.
3. Time devoted to activities that mostly benet the rm is more strongly correlated with
productivity than time devoted to activities that mostly yield private benets.
Guided by the theoretical framework, we estimate the correlation between the CEOs time
use, their total time at work, rm level measures of governance and productivity, conditional
on the size of the rm, whether it is listed and the industry it is in. The analysis yields three
key ndings, corresponding to results 1-3 above.
First, the insider/outsider allocation is associated with systematic dierences in CEO
worktime. In particular, CEOs who work longer hours spend more time with insiders and
less time, in absolute terms, with outsiders, especially in on one-on-one meetings.
Second, the insider/outsider allocation is systematically correlated with dierences in
rm governance. More precisely, we correlate CEOs time allocation with a range of external
proxies for rm level governance. We begin by comparing the time allocation of CEOs working
for domestic and multinational rms. We expect the latter to have stronger governance as
multinational rms face global competition and therefore need to provide steeper incentives
to maintain their productivity advantage (Bloom and Van Reenen 2010 and Bandiera et. al
4
2010). We then look at variables that proxy for the CEOs power vis-a-vis the rm owners.
First, we compare external CEOs who work for family rms to those hired by rms owned
by disperse shareholders. Intuitively, when ownership is concentrated, as it is in family rms,
owners might face less diculties in monitoring the CEOs time use. Second, we test whether
time allocation is correlated with board size. Large boards have long been suspected of being
dysfunctional and to lead to too little criticism of management performance, thus granting
CEOs more freedom (e.g. Lipton and Lorsch, 1992). Consistent with this view, there is
some evidence that larger boards reduce the value of the rm (e.g. Yermack, 1996). Third,
following a recent literature pointing at the fact that the presence of women on the board
might improve the quality of governance (Adams and Ferreira 2009), we use an indicator
for whether there is at least a woman in the board with an executive role. Finally, as a
more direct (and arguably exogenous) measure of governance quality and restraints on the
managers we use the country-level indicator of private benets of control computed by Dyck
and Zingales (2004) on the basis of block control transactions in 39 countries. We attach
this measure of quality of governance to multinational rms in our sample on the basis of
their country of origin and rely on within variation among multinational rms operating in
Italy. The correlations between time use and the ve independent governance proxies paint
a consistent picture: CEOs who work for rms with better governance devote more time to
insiders and less time to meeting outsiders alone.
In the nal part of our empirical analysis, we show that the insider/outsider allocation
is correlated with rm performance. Time spent with insiders is positively correlated with
several measures of rm performance, while time spent with outsiders only is not. For
example, a one percent increase in hours dedicated to insiders is correlated with a productivity
increase of 1.22%, whereas a one percent increase in hours dedicated to outsiders is correlated
with a productivity increase of 0.22%, and both eects are signicantly dierent from zero
at conventional levels.
Taken together, our ndings are consistent with the idea that dierences in time allocation
reect dierences in governance and that, compared to time spent with outsiders alone, time
spent with insiders is more benecial to the rm. Since the data at hand, however, is not
suited to identify the causal eect of governance on time allocation, we can make precise the
set of assumptions needed to reconcile the ndings with the view that dierences in time
allocation reect optimal responses to dierent shocks faced by dierent rms. The nal
section of the paper discusses these in detail.
The paper is structured as follows. After a brief literature review in Section 2, Section
3 explains our empirical methodology and describes the data. Section 4 presents a simple
time allocation model. Section 5 reports empirical ndings based on our theoretical set-up.
5
Section 6 concludes.
2 Related Literature
The interest in the use of time of top executives is not new in the management literature
4
. The
seminal work of Mintzberg (1973), based on intensive daily observation of ve managers over a
period of ve weeks, provided an in-depth characterisation of the type of activities undertaken
by executives, as well as the rst documented evidence of the extreme fragmentation and
unpredictability of their daily routines. In a similar spirit, Kotter (1982), studied fteen high
level general managers over a range of nine US corporations using three in depth interviews
conducted over a period lasting between six and twelve months, to document how the presence
of a exible agenda allows executives to gather information and build networks within the
rm. More recently, Nohria and Porter (2009), studied the daily activities of a top US CEO
over three full months. Although these studies have clearly played an important role in
providing evidence on the nature of managerial work, they are all based on very small and
selected samples of individuals, a feature that ultimately limits the generalizability of their
ndings. Furthermore, they almost exclusively focus on US based executives. Our paper
directly addresses these limitations providing, to the best of our knowledge, the rst large
scale evidence on CEO time use in an international context.
The collection and analysis of time use data has experienced a recent surge also in the
economics literature, with papers that have been mostly interested in shedding light on the
work/leisure choice of employees, and in measuring on the job welfare (see, e.g. Aguiar and
Hurst 2007, Krueger et al 2009). Closer to our paper are Hamermesh (1990) and Luthans
(1988), who analyse how workers and middle managers, respectively, spend time at work.
Hamermesh (1990) analyzes time diaries of a sample of 343 US workers to shed light whether
time devoted to leisure on-the-job (e.g. on breaks) is productive for the rm, or an indicator
of shirking. Luthans (1988) records in detail the activities of 44 managers with the aim to
identify the determinants of a managers success within his or her organization. Although
this literature has also provided a very rich set of information on the nature of managerial
activities, it has not directly tested the implications of dierent time use allocations for rm
performance, which is one of the key contributions of our paper.
Our analysis is also related to the vast literature that focuses on agency problems for CEOs
(see the survey by Stein (2003)). Our paper studies one particular form of moral hazard,
that relates to the diversion of CEOs time allocation, and hence it is quite far from other
4
See Hales (1986) for an extensive review of works focusing on the use of time of managers in the manage-
ment literature.
6
types of misbehavior studied in the literature. The most closely related study is Malmendier
and Tate (2009), who look at what happens when CEOs receive prestigious business awards,
both in terms of rm performance and CEO behavior. One of their ndings is that award
recipients appear to devote more time to writing books, playing golf, and sitting on boards
of other companies, a nding that is not inconsistent with ours in so far time to write books
crowds CEOs out working time rather than his leisure.
Finally, we see our paper as complementary to Bertrand and Schoars (2003) analysis of
management style in a panel of matched CEO-rm observations. While their objective is to
identify the xed eect of individual managers, our goal is to use a direct behavior variable
- time allocation - to make inferences on the implicit incentive structure that managers face.
3 The CEOs Diaries: Descriptive Evidence on Time Use
3.1 Sources and Sample Description
Our main data source is a time use survey, which we designed to identify the activities CEOs
engage in on a day-to-day basis. The survey eectively shadows the CEO through his PA
for every day over a one week period. The PA is asked to record real-time information on
all the CEOs activities that last 15 minutes or longer in a time use diary. Table A1 shows
how the diary is structured. For each activity - dened as a task to which the CEOs devotes
time in excess of 15 minutes - the diary records information on the type of activity (e.g.
meetings, phone calls etc), its duration, its location, whether it was scheduled in advance
and when, whether it is held regularly and how often. The diary also collects information
on the number of participants, whether these belong to the rm or not, and if insiders,
their occupational areas (e.g. nance, marketing); if outsiders, their relation to the rm
(e.g. investors, suppliers). The PA is also asked to record the total time the CEO spends
in activities that last 15 minutes or less and in transit. Hence, by summing the time spent
over activities in excess of 15 minutes and the time spent in activities that last less than 15
minutes we obtain a measure of the CEO total working time. Each PA is randomly assigned
one of ve weeks between February 10 and March 14, 2007.
The master sample contains the top 800 Italian rms from the Dun&Bradstreet data
base (2007) and the top 50 Italian banks from the list of all major Italian nancial groups
compiled yearly by the Research Division of Mediobanca, a leading Italian investment bank.
Size is measured as yearly revenue for rms and as the average of (i) employment, (ii) Stock
market capitalization; (iii) Total value of loan portfolio for banks. All 850 institutions were
contacted to ascertain the identity and contact details of the CEO; this procedure yielded 720
complete records. Of these, 50 were randomly selected for a pilot survey and the remaining
7
670 formed the nal sample. The response rate was 18%, yielding complete information on
the time use of 119 CEOs.
The implementation of the survey was outsourced to a professional survey rm, Carlo
Erminero and co., headquartered in Milan. Sample CEOs received an ocial invitation letter
from the Fondazione Rodolfo Debenedetti that sponsored this project, followed by a personal
phone call explaining the purpose of the survey and the relevant condentiality clauses. Upon
acceptance, the survey was mailed to the PA identied by the CEO, who was asked to record
the information and send back the completed forms via either fax or mail.
To maintain comparability across individuals, we drop ten CEOs who also cover the role
of "board president". The nal sample contains 94 CEOs.
5
We match the time use data with two further sources. The rst is the Amadeus data
base, which contains balance sheet data, rm size, ownership and multinational status for
94 rms in our sample. The second is the Infocamere database, which provides a rich set of
demographic information about all board members with executive and non executive respon-
sibilities.
Our dataset has obvious strengths and appeals; but it is subject also to a number of
limitations which deserve to be mentioned up-front. Apart from its small size, which reects
our focus on the largest corporations together with a low survey response rate (18%), the
data are cross-sectional in nature. We do not observe work-related activities that the CEO
may perform in the evenings or on weekends, and we miss the characteristics of those that
take less than 15 minutes.
Table 1 describes our sample rms and CEOs. Panel 1A shows that manufacturing is the
most common sector (39%) followed by Finance, Insurance and Real Estate (15%), Services
(14%), Transportation, Communication and Utilities (13%), Wholesale and Retail (8%),
Construction (4%) and Mining (3%). The median rm in our sample has 1,244 employees
and its productivity (measured as value of sales per employee) is USD 600,000 per year. Just
under 40% of the sample rms are widely held, 22% belong to families, and the remainder is
split between private individuals, private equity, government and cooperatives. Finally, the
sample 30% of the rms are domestic, 36% Italian multinationals and 34% Italian subsidiaries
of foreign multinationals. For the latter, respondents in our sample are CEOs of the Italian
5
We analyzed whether the survey sample was systematically dierent from non respondents in terms of
observable rm level characteristics, such as size, productivity, prots, location and industry. The analysis
(presented in Table A2 in appendix) shows that the CEOs participating in our study tended to work for larger
rms, and had a higher probability of being aliated to the Fondazione Debenedetti. In terms of industry
representativeness, we also had a higher response rate for rms classied in Public Administration, Business
Services and Finance. However, we could not nd any systematic dierence in terms of regional location
and rm performance, as measured by productivity, prots and three years sales growth rates. The selection
analysis is limited to the sample of 535 rms that could be matched to the Amadeus dataset
8
subsidiary.
3.2 CEOs Time Use: Insiders vs Outsiders
Table 2 and Figure 1 illustrate how CEOs spend their time. The average CEO in our sample
works 47.7 hours over a 5 day-week. We note that this only comprises the work hours known
to the PA, and as such it does not include working time at home or at the weekend. The
gure thus provides a lower bound on the working hours of the CEOs. Of these, 36.9 -just
under 80%- are spent in activities that last 15 minutes or longer, for which the diary records
detailed information. The rest of the analysis will focus on these activities that, reassuringly,
comprise the bulk of the CEOs time.
Table 2 (top panel) and Figure 1 show that, as expected, CEOs spend most of their time
(85%) with other people. Meetings take up 60% of the working hours, and the remaining
25% is comprised of phone-calls, conference calls and public events. Our sample CEOs
spend on average 15% of their time working alone. We note that this might be subject to
measurement error, as the CEO might be in his oce but not necessarily working. This is
of little consequence for our analysis, which focusses on the time the CEO spends with other
people.
We illustrate how the CEOs allocate their time across dierent categories of people, and
we focus especially on the distinction between people who belong to the rm (insiders) and
those who do not (outsiders). The diaries reveal three interesting patterns. First, while
insiders are present most of the time, CEOs also spend a considerable amount of time with
outsiders alone (Figure 1B). On average, CEOs spend 42% of their time with insiders only,
25% with both insiders and outsiders and 16% with outsiders alone. Second, time spent with
insiders is evenly spread across dierent operational areas (Table 2B). Among the top ve
categories of insiders -ranked by total time spent, the ratio between the highest (nance: 8.6
hours) and the lowest (human resources: 5.5 hours) is 1.4. In contrast, consultants dominate
among outsiders. Looking again at the top ve categories by time spent, the ratio between
the highest (consultants: 4.7 hours) and the lowest (suppliers: 1.3 hours) is 3.9. Third,
most of the variation between insiders is on the intensive margin: at least 70% of our sample
CEOs spends at least 15 minutes per week with each of the top ve categories of insiders. In
contrast, time spent with dierent categories of outsiders exhibit considerable variation on
the extensive margin.
More interestingly, we note that the average values hide a considerable amount of vari-
ation. Figure 2 shows the histograms of total diary-recorded hours, and hours spent with
insiders and outsiders. There is considerable variation in hours worked: the CEO at the
90th percentile works 20 hours longer than the CEO at the 10th percentile (47 vs 27). The
9
dierences in hours spent with various participants are even starker. The CEO at the 90th
percentile devotes 35 weekly hours to activities where there is at least one insider, whereas
the corresponding gure is 14 hours for the CEO at the 10th percentile. The CEO at the
90th percentile devotes 11 weekly hours meeting alone with outsiders, whereas the CEO at
the 10th percentile spends no time at all with outsiders alone.
To test whether these dierences are systematic rather than the outcome of random
variation across CEOs in time input requirement or purely driven by rm characteristics, we
estimate the time spent with at least one insider at the most disaggregated level of observation
in our data - the activity level - conditional on rm size, industry dummies and CEO xed
eects. We observed a total of 2,612 activities in the sample and on average, CEOs engage
in 27 activities over the observation week. We reject the null hypothesis of no systematic
dierence, namely that the CEO xed eects are jointly equal to zero, at the 1% level. The
estimated xed eects, reported in Figure A1 with their 95% condence interval, illustrate
that conditional on rm size and industry, CEOs exhibit dierent patterns of time use.
4 A Model of CEO Time Allocation
The goal of this section is to develop a minimalistic model of managerial time allocation
to illustrate how the interpretation of dierences in time allocation among dierent CEOs
depends on the extent to which the CEOs and the rms interests are aligned. We will show
that when these are perfectly aligned, time allocation is entirely determined by the rms
objectives, whereas when they are not, time allocation reects dierence in governance. In
this case the model will produce a set of testable correlations that can be tested with our
data. The points we make would still hold in a more general model, but in the interest of
space and readability we prefer to illustrate them in the simplest possible linear quadratic
formulation.
The section ends by dealing with the connection between desired time allocation, which
is the activity time distribution predicted by the model, and observed time allocation, which
is the particular realization of the theoretical time distribution that we observe in our CEO
diaries.
4.1 Denitions
The CEO faces n activities and allocates non-negative time vector (x
1
, ..., x
I
) to the activities.
The rms production function is
Y =
n

i=1

i
x
i
10
The vector describes the value of the top managers time in all possible activities and it is
determined by the rm technology and environment.
The CEO can also produce some personal rent (e.g. networking), with production function
R =
n

i=1

i
x
i
The vector depends on characteristics of the CEO and the institutional and economic
environment he operates in.
The total cost of time for the CEO is
C =
1
2
n

i=1
x
2
i
.
There is an increasing marginal cost in devoting time to one particular activity, due either
to the onset of boredom or to a lower time-eciency (for instance, take the time devoted to
having lunch with clients; the time spent to meet one client equals one hour for the meal plus
transportation time; the CEO rst meets the clients who are willing to come to the rm or
those who work nearby; then he has to spend more time traveling if he wants to meet the
ones in more distant locations).
6
The CEOs payo is
u = bY + (1 b) R C.
The parameter b captures the alignment between the rms interests and the CEOs implicit
or explicit incentive structure. If b = 1, the rm and the CEO have perfectly aligned
interests. If b = 0, the CEO only pursues personal interest.
4.2 Perfect Governance
We begin by analyzing the benchmark case where the CEOs and the rms interests are
perfectly aligned The CEO solves
max
x
n

i=1

i
x
i

1
2
n

i=1
x
2
i
6
This formulation is equivalent to one where the production function is logarithmic
Y =
n

i=1
i log xi
R =
n

i=1

i
log xi
and the cost function is linear
C =
n

i=1
xi.
11
and the solution is:
Proposition 1 In the perfect-alignment case (b = 1), the CEO devotes time to activities in
proportion to the relative value of the activities to the rm. Namely, given activities i and j,
the equilibrium allocation of time on the two activities ( x
i
and x
j
) satises:
x
i
x
j
=

i

j
.
The proposition makes clear that when the CEOs interest is perfectly aligned with the
rms, observing data on time allocation allows us to back out the relative importance of the
dierent activities in the eyes of the rm.
4.3 Imperfect Governance
When b (0, 1), the CEO pursues both the rms interest and her own. The solution to the
CEOs maximization problem no longer implies that time is allocated according to the rms
needs. Given two activities i and j, we have:
x
i
x
j
=
b
i
+ (1 b)
i
b
j
+ (1 b)
j
.
Now, with no additional information, we are unable to back out the coecients from
the allocations x for a particular CEO. However, we can derive a number of cross-sectional
implications that will guide our empirical work.
To put some structure on the problem, We make two assumptions. First, we assume that
activities can be grouped into two sets: I
Y
and I
R
. The rst set contains activities that
benet the rm more than the CEOs (
i
>
i
when i I
Y
), while the second set contains
activities that yield more personal benet (
i
<
i
when i I
R
). Furthermore, activities in
I
Y
are better for the rm: if i I
Y
and j I
R
, then
i
>
j
(otherwise there are activities
that are dominated). In particular, our assumption is satised when activities can be split
into two groups: tasks that only benet the rm (
i
> 0 and
i
= 0) and tasks that only
benet the CEO (
i
= 0 and
i
> 0). As shorthand, we call activities in I
Y
benecial to
the rm and activities in I
R
benecial to the CEO, but recall that this denition should
be preceded by comparatively more.
Second, we assume that the benet to the rm from Y activities is greater in monetary
terms than the benet to the CEO of R activities, namely

kI
Y

k
>

kI
R

k
12
This is a weak condition; its violation would imply that the CEO job exists mainly for the
benet of the job holder rather than for the rm.
There is a distribution of rms that are identical except for the quality of governance, so
that b is a random variable with support (0, 1). If we observe a distribution of realized time
allocations (a vector x for each rm), we can state
Proposition 2 In equilibrium, the cross-sectional correlation between the time x
i
that the
CEO devotes to a particular activity and the total time the CEO spends at work is positive
for activities that are benecial to the rm and negative for activities that are benecial to
the CEO.
Corr

k=1
x
k
, x
i

> 0 i I
Y
Proof. Recall that the equilibrium allocation is given by x
i
= b
i
+(1 b)
i
. The covariance
whose sign is the same as the correlation is
Cov

k=1
x
k
, x
i

= Cov

k=1
(b
k
+ (1 b)
k
) , b
i
+ (1 b)
i

= V ar (b)

i
n

k=1

i
n

k=1

i
n

k=1

k
+
i
n

k=1

= V ar (b) (
i

i
)

k=1

k=1

which is positive if and only if

i
>
i
.
The intuition behind proposition 2 has to with better discipline. Recall that rms dier
in terms of governance quality b. Firms with a higher b induce the CEO to spend more
time on Y -tasks that benet the rm (relatively more) and less on R-tasks that benet the
CEO. This determines an increase and a decrease in total work time. However, the second
assumption guarantees that the former eect is strictly larger. Thats because there is more
potential benet to be gained through activities that are benecial to the rm. If the rm
owners are able to translate this into monetary incentives for the CEO, they can obtain a
higher level of eort than the CEO is willing to put on activities that are directly benecial
to herself.
Second, suppose we have a, possibly noisy, measure of governance. Let

b = b + ,
where

b is the governance proxy and is an i.i.d. noise term. Note that

b could be a poor
proxy, in the sense that the variance of can be very high.
13
Unsurprisingly, better governance is positively related to activities that are more benecial
to the rm and negatively related to R activities:
Proposition 3 The governance measure

b is positively correlated with time spent on an
activity if and only that activity is benecial to the rm.
Cov

b, x
i

> 0 i I
Y
;
Proof. We have
Cov

b, x
i

= Cov (b + , b
i
+ (1 b)
i
) = V ar (b) (
i

i
) ,
whose sign depends once more on whether i is a productive activity.
Finally, we can relate time allocation to rm productivity Y .
Proposition 4 In equilibrium, the cross-sectional correlation between the time x
i
that the
CEO devotes to an activity i and rms productivity

Y is positive if and only if activity i is
benecial to the rm. Namely
Corr

Y , x
i

> 0 i I
Y
;
Proof. Recall that
Y =
n

k=1

k
x
k
=

kI
Y

k
x
k
We have
Cov

Y , x
i

= Cov

k
(b
k
+ (1 b)
k
) , b
i
+ (1 b)
i

= Cov

b
2
k
+ (1 b)
k

, b
i
+ (1 b)
i

= V ar (b)

i
n

k=1

2
k

i
n

k=1

2
k

k
+
i

= V ar (b) (
i

i
)

k=1

k
(
k

k
)

which again is positive if and only

i
>
i
, because there exists such that
i
>
whenever i I
Y
and j I
R
and hence

n
k=1

k
(
k

k
)

n
k=1
(
k

k
) 0.
The usefulness of the three previous results is that they allow us to identify the ralative
value of an activity to the rm and the CEO. If we do not know a priori which activities are
more benecial to the rm and which activities are more benecial to the CEOs, the three
propositions oer three distinct ways of using time allocation information to distinguish
between these. In the empirical application we use this framework to shed light on whether
time with insiders and outsiders can be divided along these lines.
14
5 Evidence on CEOs Time Use
Guided by the theoretical framework we now explore the correlation between the time use
of the CEO and three sets of variables: (i) CEOs eort, (ii) rm governance and (iii) rm
performance. It is key to stress that our analysis only uncovers equilibrium correlations, that
is the nature of the data does not allow us to establish causal links. The analysis will however
shed light on the set of assumptions under which the evidence can be made consistent with
the two views of the world -perfect vs imperfect governance- formalized by the theoretical
framework. We will discuss these assumptions in detail in Section 6.
5.1 CEOs Time Use and Hours Worked
The descriptive evidence in Figure 2 indicates that CEOs dier considerably both in the
amount of time they work and in how they allocate this across dierent categories of people.
Our rst test aims at assessing whether eort -measured by hours worked- and time use are
correlated. Namely do CEOs who work longer hours have a systematically dierent pattern
of time allocation between insiders and outsiders?
Throughout we focus on the distinction between time spent with at least one insider- with
or without outsiders- and time spent with outsiders alone. This follows from the reasonable
assumption that if the CEO devotes some of his time to activities that only carry private
benets, he is more likely to do so when no other rm employee is present.
Table 3 estimates the correlations between hours worked and time allocated to insiders
and outsiders alone. Throughout we control for rm size, whether the rm is listed and
industry dummies (coecients not reported), as the time CEOs must devote to outsiders
is likely to be determined by these. Panel A shows that there is a signicant correlation
between the total time the CEO spends at work and how he allocates this between dierent
categories of people. In particular, hours devoted to insiders increase one to one with hours
worked, whereas hours spent with outsiders alone are negatively correlated with total hours
worked. The table reports the p-value of the test of the null hypothesis that the coecient is
equal to one- namely that hours with insiders (outsiders) increase in direct proportion to the
hours worked. We nd that the coecient in the regression of the logarithm of hours spent
with insiders on the logarithm of total hours is not statistically dierent from one, whereas
the coecient in the corresponding regression for hours spent with outsiders only is negative,
and signicantly dierent from one.
Panel B shows that the negative correlation between hours worked and hours spent with
outsiders alone is driven exclusively by the fact that CEOs who work longer hours devote
fewer hours to one-to-one meetings with outsiders alone. Columns 3 to 6 divide the time spent
15
with insiders and outsiders into time spent with one or more categories of each. For instance,
a meeting with investors would fall under the "one category of outsiders" heading, whereas
a meeting with investors and consultants would fall under the "more than one category of
outsiders" heading. Columns 3 and 4 show that the correlation between hours worked and
time spent with insiders does not depend on whether these belong to one or more categories.
In contrast, Columns 5 and 6 show that CEOs who work longer hours spend fewer hours in
meetings alone with one type of outsiders only. The correlation with hours spent with more
than one category of outsiders is positive but signicantly less than one.
Panel C divides the time spent with insiders and outsiders according to the number of
people participating in the activity. In line with the ndings above, Columns 7 and 8 show
that the correlation between hours worked and time spent with insiders does not depend on
the number of insiders present. In contrast, Column 9 shows that CEOs who work longer
hours spend fewer hours in meetings with one or two outsiders and nobody else.
5.2 CEOs Time Use and Governance
Findings so far indicate that CEOs who work longer hours devote more time to insiders and
less time- in absolute terms- to meeting outsiders alone, especially in one-to-one meetings. To
the extent that CEOs whose rms have better governance work harder, this is consistent with
the interpretation that observed variation in time allocation reect dierences in governance.
We can test this relationship directly, by analyzing the correlation between time use and ve
independent proxies of "governance quality". These proxies rely on the intuition that the
CEOs ability to pursue activities that only produce private benets, depends on the strength
and eective functioning of the rm board and its ability and willingness to monitor CEOs
actions, as well as on the economic cost the CEO pays if the rm underperforms.
In this framework, governance quality is both the reection of economic incentives and
the boards ability to monitor the CEO. Since misalignment between the rm and the CEO
objectives is not directly observable, we proxy quality of governance using ve external rm
level indicators. First we use an indicator of whether the rm is a multinational. Intuitively,
because of the stronger competitive pressure that multinationals face, they need keep CEOs
focused on rm productivity to preserve their competitive advantage. In line with this Bloom
and Van Reenen (2010) and Bandiera et. al (2010) show that multinational rms are more
likely to be well managed and to oer steep performance incentives to their executives. Hence
we would expect that CEOs of multinational rms incur a higher cost if they divert time
from productive to unproductive uses. Supporting this idea, we nd that CEOs who work
for multinationals spend more time with insiders and less time with outsiders alone. In
particular, Columns 1 and 2 in Table 4 show that CEOs who work for foreign multinationals
16
spend 54% more time with insiders and 55% less with outsiders alone. Both coecients are
precisely estimated at conventional levels. Time allocation in Italian multinationals is similar
and we cannot reject the null that the coecients are equal.
The next set of indicators proxies for the power of rm owners vis--vis the CEO. The
rst indicator in this set measures whether the rm is owned by a family, and if so, whether
the CEO is external to the family or one of its members. This distinction is key because
it raises dierent governance issues. Compared to disperse shareholders, family owners face
fewer diculties in monitoring the CEOs they hire to run their rms, for instance because
coordination among owners is much easier when owners are few. When the CEO belongs to
the family, there is no separation of ownership and control and hence no governance issues
as such; however, family CEOs in family rms might pursue objectives other than prot
maximization (Bandiera et al 2010) and the implication of these for time allocation is not
captured by our simple framework. Columns 3 and 4 in Table 4 show that external CEOs who
work for family rms spend 32% more time with insiders and 27% less time with outsiders
alone, but only the former is estimated precisely. This is in line with the intuition that,
compared to disperse shareholders, family owners can keep a tighter control on their CEOs.
Family CEOs, in contrast, do not appear to behave dierently than CEOs in non family
rms.
The second indicator is the size of the board. Large boards have long been suspected
of being dysfunctional and to provide too little criticism of management performance, thus
granting CEOs more discretion which can be abused (e.g. Lipton and Lorsch, 1992). Jensen
(1993), subscribing to this criticism argues that "when boards get beyond seven or eight
people they are less likely to function eectively and are easier for the CEO to control". In
addition, pletoric boards can be just the reection of attempts to bring in friends and transfer
them shareholders money through board members compensation. Consistent with this views
there is some evidence that larger boards reduce the value of the rm (e.g. Yermack, 1996).
In line with this, Columns 7 and 8 show that CEOs with larger boards spend less time with
insiders and more time with outsiders alone. A 1% percentage point increase in board size
reduces the time spent with insiders by 0.24 of a percentage point while it increases the time
spent with outsiders alone by 0.47 of a percentage point.
A better measure of constraints on the CEO is the independence of board members. We
have no direct information on the nature of the directors. But we observe some of their
characteristics, including gender. As a third proxy for quality of governance we use an
indicator for whether there is at least a woman in the board. There is evidence that the
presence of women improves the quality of governance. For instance, Adams and Ferreira
(2009) nd that boards with more women tend to be tougher with the CEO, to raise board
17
attendance and to monitor more.
7
Throughout the analysis, we distinguish between executive
and non executive roles, since we expect this eect to be stronger when women assume
executive roles. The ndings in columns 9 and 10 indicate that CEOs spend 17% more time
with insiders and 27% less with outsiders alone, respectively when at least one woman on the
board has an executive role. As expected, the correlations have a similar sign but are smaller
in magnitude and not precisely estimated when we look at the presence of women with non
executive roles.
As a nal and more direct measure of governance quality and restraints on the man-
agers we use the country-level indicator of private benets of control computed by Dyck and
Zingales (2004) on the basis of block control transactions in 39 countries. We attach this
measure of quality of governance to multinational rms in our sample on the basis of their
country of origin and rely on within variation among multinational rms operating in Italy.
Compared to the others, the main advantages of this measure are its direct interpretability in
terms of access to private benets and its exogeneity. It comes at the cost that we can only
rely on a subset of observations. The ndings in columns 11 and 12, table 4, indicate that
CEOs who work for multinationals based in countries with better governance spend more
time with insiders and less with outsiders alone. A one standard deviation increase in the
index, indicating a worsening of governance, decreases hours spent with insiders by 13% and
increases hours spent with outsiders alone by 45%.
Taken together, the ndings in Table 4 provide a coherent picture. Whenever the CEO
and the rm interest are better aligned - either because of stronger market competition,
better monitoring, more powerful boards- CEOs spend more time with insiders and less time
with outsiders alone. This, as well as the ndings in the previous section, suggests that
the evidence is consistent with the idea that time with outsiders alone yields more private
benets than time with insiders, and that dierences in time allocation reect dierences
in governance. The next section looks at whether dierences in time allocation have any
implication for rm performance.
5.3 CEOs Time Use and Firm Performance
For our next test we match the time use data with external measures of rm performance
from the Amadeus database. Following the theory, we test whether the time the CEO spends
7
One rational is that women have higher standards of morality and a higher sense of their duty as docu-
mented in experimental and survey based studies which show that women are more likely to stick to the rules,
less likely to break agreed legal or moral norms (e.g. Guiso et. al (2003) ) and are less self-oriented then men
(for instance Eagly and Crowley (1986); Eckel and Grossman (1998); Reiss and Mitra (1998); Butler et. al.
(2010)). Because of this they are more willing to voice in the board meetings against a shirking CEO, thus
providing more monitoring and control.
18
with dierent categories of people is correlated with rm performance. As illustrated by the
model in Section 3, all hours devoted to productive activities should be positively correlated
with rm performance, namely the CEOs time is an input in the production function. If,
on the other hand, the CEO spends time on activities that benet him but not the rm, the
time devoted to these activities should not be correlated to rm performance.
Table 5 estimates the conditional correlation between rm performance and CEOs time
use controlling for rm size and industry dummies. Our main measure of performance is
productivity, measured as the value of sales over employees, which is available for 80 of
our 94 sample rms. Firms in the nance sector are excluded from this exercise because
sales/employees is not a comparable measure of productivity in this sector.
Column 1 shows a strong positive correlation between hours worked and productivity.
The coecient is precisely estimated and implies that a one percentage point increase in
CEOs working hours is associated with a 2.14 percentage point increase in productivity.
Column 2 shows that this correlation is entirely driven by the hours the CEOs spend with
at least one other employee of the rm. A one percentage point increase in hours spent with
at least one insider is associated with a 1.23 percentage points increase in productivity. In
contrast, hours spent working alone and, most interestingly, hours spent with outsiders alone
are not correlated with productivity. The coecient is precisely estimated but small and not
signicantly dierent from zero. The test of equality of coecients between hours with at
least one insiders and hours with outsiders alone rejects the null at the 1% level.
Columns 3 to 5 address measurement error issues. In particular, given that the CEO
generally meets insiders at the rms HQ, while he might meet outsiders in other locations,
the results could be driven by measurement error if the time spent with outsiders comprises
unproductive travel or setting-up time. Alternatively, the PA might have more precise in-
formation on the hours dedicated to a given activity if this takes place in the CEOs oce,
which is presumably close to hers, rather than on activities that take place away from the
rm. To allay this concern, Column 3 splits the hours spent with outsiders alone according
to location. If this were driving the results, we would expect the time spent with outsiders at
HQ to be positively correlated with rm productivity. The results indicate that however this
is not the case. The coecient on hours spent with outsiders is small, precisely estimated
and not signicantly dierent from zero regardless of where the activity takes place.
A related concern is that the PA might have more precise information about the duration
of activities with insiders and hence measurement error would again introduce a downward
bias on the coecient on time with outsiders alone. We address this concern using infor-
mation on whether the activity was scheduled in advance and recorded in the CEOs diary.
Intuitively, activities that are planned in advance and already recorded should be measured
19
more precisely. Columns 4 and 5 show that this form of measurement error was not driving
the results either. The coecient of hours spent with outsiders is small, precisely estimated
and not signicantly dierent from zero regardless of whether the activity was planned in
advance. In contrast, the coecient on hours spent with at least one insider is positive and
signicantly dierent from zero regardless of whether the activity was planned in advance.
Column 5 probes the robustness of the results to the inclusion of other inputs, namely capital
and materials.
While the pattern of time use is correlated with productivity, it might also be correlated
with other unobservables - most notably the CEOs pay - so to leave the rm owners indif-
ferent between having a hard working CEO who spends long hours with insiders and a CEO
who works fewer hours and spends more time with outsiders alone. To assess the empirical
relevance of this interpretation, Column 6 presents the conditional correlation between rms
prots per employee and time use. The results show that time spent with at least one insider
is positively correlated with prots, whereas time spent with outsiders is not.
A possible further concern is that the impact of time allocation choices might reveal
itself over a longer time period. For example, networking activities might need a longer
time frame to bear their positive impact on rm performance. To test this idea we consider
as a dependent variable the growth rate of sales over a three year period. The results in
column 7 do not support the idea that time spent with outsiders might have a positive long
run impact. In fact, although the coecients are not precisely estimated, time spent with
outsiders is negatively correlated with three-year sales growth, while time spent with insiders
shows a positive correlation.
6 Discussion and Conclusions
The analysis above has illustrated a rich set of correlations between the CEOs pattern of time
use, his eort, governance proxies and the performance of the rm he leads. In particular we
nd that CEOs who work longer hours spend more time in activities that involve at least one
insider and less time in activities where the CEO meets outsiders alone. CEOs who work for
rms with stronger governance also spend more time in activities that involve at least one
insider and less time in activities where the CEO meets outsiders alone Finally, time spent
with insiders is positively correlated with productivity and prots, whereas time spent with
outsiders alone is not.
As illustrated in the theoretical framework, variations in time use can either be seen as
optimal responses to dierent circumstances faced by dierent rms if the CEOs ad the
rms interests are perfectly aligned (b = 1) or as consequences of dierences in the quality of
20
governance. We now make precise the conditions under which the ndings can be reconciled
with either the perfect and the imperfect governance view. It is important to stress that our
data does not allow us to implement a formal test of the two models, but the rich set of
correlations we uncover allows us to pin down the set of assumptions needed to reconcile the
ndings with either view.
In the perfect governance view, where the CEOs and the rms interests are perfectly
aligned, the observed correlations must be due to shocks that aect the pattern of time use,
total hours worked, rm performance and rm characteristics as follows. First, shocks that
increase the marginal return to CEOs eort - so that CEOs work longer hours, also increase
the marginal product of time with at least one insider and at the same time decrease the
marginal product of time alone with outsiders, especially so for time alone in one-to-one
meetings. Second, shocks that increase the rms productivity, also increase the marginal
product of time with at least one insider and leave the marginal product of time alone with
outsiders constant. Third, this type of shocks are more likely to hit multinationals, family
rms with external CEOs, rms with small boards, rms with women covering executive roles
on the board, and rms from countries with a good governance record.
In the imperfect governance view, the observed variation in time use can derive from
dierences in governance that determine the extent to which the CEOs interests are aligned
with the rm. If we take this view, the ndings indicate that time spent with insiders
contributes to the rms output while time spent alone with outsiders only benets the
CEO. Moreover, CEOs hired by rms with better governance work longer hours and devote
more time to activities that benet the rm and less time to activities that mostly carry
private benets.
We leave it to the reader to assess which of the two interpretations is more plausible. In
either case, our analysis suggests that CEO time use data are a useful tool to shed light on
constraints and objectives of rms and of their executives.
References
[1] Adams, Rene B. and Ferreira, Daniel. 2009. "Women in the Boardroom and Their
Impact on Governance and Performance." Journal of Financial Economics, 94 (2):291-
309.
[2] Aguiar, Mark and Erik Hurst. 2007. Measuring Trends in Leisure: The Allocation of
Time Over Five Decades 2007. Quarterly Journal of Economics, August 2007, 122(3),
969-1006
21
[3] Bandiera, Oriana; Guiso, Luigi; Prat, Andrea and Sadun, Raaella. 2010. "Matching
Firms, Managers, and Incentives." Harvard Business School Working Paper 10-073.
http://www.hbs.edu/research/pdf/10-073.pdf.
[4] Bennedsen, Morten; Nielsen, Kasper Meisner; Perez-Gonzalez, Francisco and Wolfenzon,
Daniel. 2007. Inside the Family Firm: The Role of Families in Succession Decisions and
Performance. Quarterly Journal of Economics, 122:647-91.
[5] Bertrand, Marianne and Schoar, Antoinette. 2003. "Managing with Style: The Eect of
Managers on Firm Policies." Quarterly Journal of Economics, 118 (4):1169-208.
[6] Bloom, Nicholas and Van Reenen, John. 2010. "Why Do Management Practices Dier
across Firms and Countries?" Journal of Economic Perspectives, 24 (1):203-24.
[7] Bolton, Patrick and Dewatripont, Mathias. 1994. "The Firm as a Communication Net-
work." The Quarterly Journal of Economics, 109 (4):809-39.
[8] Butler, Je ; Giuliano, Paola and Guiso, Luigi. 2010. "Trust and Cheating." UCLA
mimeo.
[9] Drucker, Peter F. 1966. The Eective Executive. New York,: Harper & Row.
[10] Drucker, Peter F.; DAntonio, William V. and Ehrlich, Howard J. 1961. Power and
Democracy in America. Notre Dame, IN: University of Notre Dame Press.
[11] Dyck, Alexander and Zingales, Luigi. 2004. "Private Benets of Control: An Interna-
tional Comparison." The Journal of Finance, 59 (2):537-600.
[12] Eagly, A., and M. Crowley. 1986. "Gender and Helping Behavior: A Meta-analytic View
of the Social Psychological Literature." Psychological Bulletin 100:283-308.
[13] Eckel, Catherine C. and Grossman, Philip J. 1998. "Are Women Less Selsh Than Men?:
Evidence from Dictator Experiments." Economic Journal, 108 (448):726-35.
[14] Garicano, Luis. 2000. "Hierarchies and the Organization of Knowledge in Production."
Journal of Political Economy, 108 (5):874-904.
[15] Guiso, Luigi; Sapienza, Paola and Zingales, Luigi. 2003. "Peoples Opium? Religion and
Economic Attitudes." Journal of Monetary Economics, 50 (1):225-82.
[16] Hales, Colin P. 1986. "What Do Managers Do? A Critical Review of The Evidence."
Journal of Management Studies 23 (1):88-115.
22
[17] Hamermesh Daniel.1990. "Shirking or Productive Schmoozing: Wages and the Alloca-
tion of Time at Work," Industrial and Labor Relations Review, 43 (3):121-S-133-S.
[18] Jensen, Michael C. 1993. "The Modern Industrial Revolution, Exit, and the Failure of
Internal Control Systems." The Journal of Finance, 48 (3):831-80.
[19] Khurana, Rakesh. 2002. Searching for a Corporate Savior: The Irrational Quest for
Charismatic CEOs. Princeton, NJ: Princeton University Press.
[20] Kotter, John P."What Eective General Managers Really Do." Harvard Business Re-
view, March-April 1999, p. 145-159.
[21] Krueger, Alan B. and Andreas Mueller. 2010. "Job Search and Unemployment Insurance:
New Evidence from Time Use Data," Journal of Public Economics, forthcoming.
[22] Krueger, Alan B., Daniel Kahneman; David Schkade; Norbert Schwarz and Arthur A.
Stone. 2009. "National Time Accounting: The Currency of Life," Measuring the Subjec-
tive Well-Being of Nations, University of Chicago Press for NBER, 2009, pp. 9-86.
[23] Laey, A. G. 2009. "What Only the CEO Can Do." Harvard Business Review, 87 (5):54-
62.
[24] Lipton, Martin and Lorsch, Jay W. 1992. "A Modest Proposal for Improved Corporate
Governance." Business Lawyer, 48 (1):59-77.
[25] Luthans, Fred. 1988. "Successful vs. Eective Real Managers." The Academy of Man-
agement Executive (1987-1989), 2 (2):127-32.
[26] Malmendier, Ulrike and Tate, Georey. 2009. "Superstar CEOs." Quarterly Journal of
Economics, 124 (4):1593-638.
[27] Mintzberg, Henry. 1973. The nature of managerial work. New York,: Harper & Row.
[28] Prez-Gonzlez, Francisco. 2006. Inherited Control and Firm Performance. American
Economic Review, 96:1559-88.
[29] Porter, Michael E., and Nitin Nohria. 2010 "What Is Leadership?: The CEOs Role in
Large, Complex Organizations." in Handbook of Leadership Theory and Practice, edited
by Nitin Nohria and Rakesh Khurana, Chapter 16, pp. 433-73. Boston, MA: Harvard
Business Press.
[30] Reiss, Michelle C. and Mitra, Kaushik. 1998. "The Eects of Individual Dierence Fac-
tors on the Acceptability of Ethical and Unethical Workplace Behaviors." Journal of
Business Ethics, 17 (14):1581-93.
23
[31] Stein, Jeremy C. 2003. "Agency, Information and Corporate Investment" in Handbook
of the Economics of Finance, eds. G. M. Constantinides, M. Harris and R. M. Stulz,
109-63. Amsterdam; Boston: Elsevier/North-Holland.
[32] Yermack, David. 1996. "Higher Market Valuation of Companies with a Small Board of
Directors." Journal of Financial Economics, 40 (2):185-211.
[33] Van Zandt, Timothy. 1998. "The Scheduling and Organization of Periodic Associative
Computation: Ecient Networks." Review of Economic Design, 3 (2):93-127.
24
!"#$%&'()&*+,-./0$,0
!"#$%&'()*+,-(.(/)+(.+).,**('(0+&1+,/,+,(2
!3 $% * +, + ,+% .,** + + , * 4
mean of sharetime_lunch
mean of sharetime_publicevent
mean of sharetime_phonecalls
mean of sharetime_workalone
mean of sharetime_meetings
0 .2 .4 .6
-((+,052
6)'72 &4)0(
8%)0( 1&442
89:4,1 (/(0+2
:92,0(22 4901%(2
!3"$%&'()*+,-(28(0+6,+%.,**('(0+1&+(5)',(2)*8()84(
mean of sharetime_lunch
mean of sharetime_publicevent
mean of sharetime_phonecalls
mean of sharetime_workalone
mean of sharetime_meetings
0 .2 .4 .6
-((+,052
6)'72 &4)0(
8%)0( 1&442
89:4,1 (/(0+2
:92,0(22 4901%(2
,02,.('2&0.)9+2,.('2
,02,.('2&0.)9+2,.('2
mean of so
mean of sharetime_mix
mean of si
0 .1 .2 .3 .4
)9+2,.('2&4)0(
,02,.('2)04;
,02,.('2&0.)9+2,.('2
!"#$%&'()&*+,-./0$,0
!"#$%&'()%'*+,
!-#$%&'((.+/01203%&0(2,+'(45%/+
1
.
2
.
3
.
4
F
r
a
c
t
i
o
n
0
.
0
5
.
1
.
1
5
.
2
.
2
5
F
r
a
c
t
i
o
n
20 30 40 50 60
weekly hrs all activities
0%0451++*563%&'(
!7#$%&'((.+/01203405+4(0%/+2/(2,+'
0
.
1
.
2
.
3
.
4
F
r
a
c
t
i
o
n
0 5 10 15 20
weekly hrs spent in this activity
0
.
0
5
.
1
.
1
5
.
2
.
2
5
F
r
a
c
t
i
o
n
0 10 20 30 40 50
weekly hrs spent in this activity
0
.
0
5
.
1
.
1
5
.
2
.
2
5
F
r
a
c
t
i
o
n
20 30 40 50 60
weekly hrs all activities
0%0451++*563%&'(
1++*563%&'(1203%&0(2,+'( 45%/+
1++*563%&'( 1203405+4(0%/+2/(2,+'
8%0+9)++*56 3%&'(4'+:%;.&0+,4(03+(&;%<02;+(.+/02/4554:02=202+(5%/>+'034/
<2<0++/;2/&0+(#)++*563%&'(1203%&0(2,+'(45%/+2(03+(&;%<02;+(.+/02/4554:02=202+(
13+'+03+7?@14(45%/+1203%&0(2,+'(<%'5%/>+'034/<2<0++/;2/&0+(#)++*563%&'(
1203%&0(2,+'(45%/+2(03+(&;%<02;+(.+/02/4554:02=202+(13+'+03+7?@14(120340
!"#$%&'"()$%*%+,-.)/.0%+
Mean Medlan SLd.uev. Cbs
Slze(numberofLmplovees) 8330 1244 30014 94
roducLlvlLv(Sales/Lmploveesx1000uSu) 4234 600 13664 80
SecLor:() 98
!"#$%"&'$()#* 39.4
+,(-)&,. 13.8
/)#"#&,01#.$("#&,02,"34.'"', 14.9
5("#.67('"')7#"#897::$#)&"')7# 12.8
;<73,."3,"#82,'")3 8.3
!)#)#* 3.2
97#.'($&')7# 4.2
Cwnershlp() 94
;)8,3=>,38 37.2
/":)3= 22.3
+'"', 18.1
?'<,( 13.3
naLlonallLv() 98
@7:,.')& 29.9
1'"3)"#!A4 36.1 .
/7(,)*#!A4 34
llrmlsllsLed 0.19 94
8oardslze 13.1 10 8.1 91
ALleasLonewomanonLheboardlnanexecuLlverole 0.36 91
ALleasLonewomanonLheboardlnanonexecuLlverole 0.68 91
rlvaLebeneflLlndex 0.023 0.038 0.049 30
.
*"/" '12-,%+ "34 5"-."#$% *%6.3./.13+7 uaLa on flrm slze, producLlvlLv and secLor are drawn from AMAuLuS. roducLlvlLv ls
noL avallable for flrms ln Lhe flnanclal secLor. Cwnershlp and mulLlnaLlonal sLaLus has been obLalned Lhrouah manual
search. 8oard daLa ls drawn from Lhe lnfocamere daLaseL. AccounLlna, ownershlp and board lnformaLlon are relaLlve Lo
2007. 1he prlvaLe beneflL lndex ls measured (onlv for non lLallan mulLlnaLlonals) aL Lhe counLrv of orlaln level and lL ls drawn
from uvck and Zlnaales (2004), 1able v, col 2.
!"#$%&'(%)*+!,-./*.
!"#$%&#
'(%$#)*+,-./01(
($.23
4)1%56)7$.
89:9
;<:<=>
4)1%56)7$.?#@)$A#A
BC:<8
;=:BB>
D)$E0F&%5)F# G:CC :<H
;8:CG>
D01(%15#%.1)F#0F.0A#$ I8:I H
;<:BG>
D01()71.0A#$.)F5J 8:== :=8
;8:=B>
01*,2.3*4!567(89.:53,.*'
K0F%F@# =:C< :9<
;C:<C>
L%$E#10F& =:BI :9B
;9:CI>
MNO0$#@1)$. C:=G :9H
;9:H=> ; >
'1$%1#&J G:=9 :C<
;G:<9>
67P%F?#.)7$@#. G:89 :98
;G:=C>
);9*,2.3*4!567(89.:53,.*
+50#F1. I:G< :83
;8:3B>
'7QQ50#$. H:I9 :B8
;I:H<>
RF"#.1)$. H:98 :BB
;B:3G>
+)F.751%F1. 8:93 :C9
;8:=I>
M%FE. H:B= :B3
;I:C=>
<898 =5;3>.* 812 ?83,8@A. <.B,1,9,51*' !55 "%$0%S5#. %$# A$%/F *$)P 1(# 10P# 7.# .7$"#J: 4)1%5
()7$. 0F@57A# %55 %@10"010#.T 0F@57A0F& 1().# .()$1#$ 1(%F HG P0F71#. %FA 1$%"#5: 4)1%5 ()7$.
$#@)$A#A )F5J @)F1%0F. %@10"010#. 1(%1 %$# 5)F&#$ 1(%F HG P0F71#. %FA A)#. F)1 0F@57A# 1$%"#5: 4(#
*0$.1 @)57PF $#Q)$1. %"#$%&# ()7$. .Q#F1 SJ 1(# .%PQ5# +,-. )"#$ 1(# $#5#"%10)F /##E /01( #%@(
)* 1(# 50.1#A @%1#&)$0#. 0F %@10"010#. 5)F&#$ 1(%F HG P0F71#.: 4(# .#@)FA @)57PF $#Q)$1. 1(# .(%$#
)* +,-. 1(%1 .Q#FA %1 5#%.1 HG P0F71#. /01( #%@( )* 1(# 50.1#A @%1#&)$0#.:
1A8LL3.CLCs'1|meUseandnoursWorked
ane|AIns|dersandCuts|ders
1 2
loa(LoLalhours)spenL: wlLhaLleasLonelnslder wlLhouLsldersonlv
loa(LoLalhours) 1.179*** 0.693**
(0.147) (0.337)
P0:loa(LoLalhours)=1 .23 .00
Ad[usLed8squared 0.473 0.207
ConLrols slze,llsLed slze,llsLed
lndusLrvuummles ves ves
CbservaLlons 94 94
ane|8Ins|dersandCuts|ders,bynumberofcategor|es|nvo|ved
3 4 3 6
loa(LoLalhours)spenL:
wlLhonecaLeaorvof
lnslders
wlLhmoreLhanone
caLeaorvoflnslders
wlLhonecaLeaorvof
ouslders
wlLhmoreLhanone
caLeaorvofouLslders
loa(LoLalhours) 0.909*** 1.082** 0.926*** 0.381*
(0.260) (0.448) (0.317) (0.223)
loa(numberofemplovees) 0.024 0.028 0.033 0.029
(0.038) (0.034) (0.046) (0.027)
llsLedflrm(=1lfves) 0.164 0.631** 0.906*** 0.146
(0.233) (0.287) (0.214) (0.202)
P0:loa(LoLalhours)=1 .73 .83 .00 .00
d d d Ad[usLed8squared 0.138 0.233 0.203 0.166
ConLrols slze,llsLed slze,llsLed slze,llsLed slze,llsLed
lndusLrvuummles ves ves ves ves
CbservaLlons 94 94 94 94
ane|CIns|dersandCuts|ders,bynumberofpart|c|pants|nvo|ved
7 8 9 10
loa(LoLalhours)spenL:
wlLhfewerLhanLwo
parLlclpanLsandaLleasL
onelnslder
wlLhmoreLhanLwo
parLlclpanLsandaL
leasLonelnslder
wlLhouLsldersonlvand
fewerLhanLwo
parLlclpanLs
wlLhouLsldersonlv
andmoreLhanLwo
parLlclpanLs
loa(LoLalhours) 0.394* 1.387*** 0.789*** 0.073
(0.303) (0.224) (0.276) (0.363)
loa(numberofemplovees) 0.010 0.027 0.040 0.013
(0.040) (0.032) (0.037) (0.039)
llsLedflrm(=1lfves) 0.082 0.182 0.366** 0.640**
(0.201) (0.130) (0.231) (0.248)
P0:loa(LoLalhours)=1 .18 .08 .00 .00
Ad[usLed8squared 0.136 0.422 0.168 0.142
ConLrols slze,llsLed slze,llsLed slze,llsLed slze,llsLed
lndusLrvuummles ves ves ves ves
CbservaLlons 94 94 94 94
Notes:*,**,***lndlcaLe LhaLLhecoefflclenLlsslanlflcanLlvdlfferenLfromzeroaLLhe10,3,1level,respecLlvelv.CLSLsLlmaLes,robusL
sLandarderrorslnparenLheses.AllrearesslonslncludelndusLrvdummles,LheloaarlLhmofLhenumberofemploveesandadummvequalLo1
lfLheflrmlsllsLed.
!"#$%&'(%)*+!,-./*.012345.61017.
1 2 3 4 3 6 7 8 9 10 11 12
dependenLvarlablelsloa(LoLalhours)spenLwlLh:
wlLhaL
leasLone
lnslder
ouLslders
onlv
aLleasL
onelnslder
ouLslders
onlv
aLleasL
onelnslder
ouLslders
onlv
aLleasL
onelnslder
ouLslders
onlv
aLleasL
onelnslder
ouLslders
onlv
aLleasL
onelnslder
ouLslders
onlv
lorelanMnL 0.428*** 0.394** 0.438*** 0.612**
(0.123) (0.237) (0.123) (0.260)
uomesLlcMnL 0.288** 0.233 0.217* 0.122
(0.122) (0.227) (0.124) (0.233)
lamllvownedLxLernalCLC 0.277** 0.337 0.313** 0.486
(0.123) (0.293) (0.133) (0.301)
lamllvownedlamllvCLC 0.113 0.133 0.072 0.046
(0.106) (0.272) (0.118) (0.288)
Loa(boardslze) 0.243* 0.474**
(0.133) (0.230)
ALleasLonewomanonLheboardlnanexecuLlverole 0.163* 0.328*
(0.096) (0.190)
ALleasLonewomanonLheboardlnanonexecuLlverole 0.032 0.230
(0.108) (0.241)
rlvaLebeneflLlndex 2.603* 9.093*
(1.132) (4.266)
lndusLrvuummles ves ves ves ves ves ves ves ves ves ves ves ves
llrmSlze ves ves ves ves ves ves ves ves ves ves ves ves
llrmLlsLed ves ves ves ves ves ves ves ves ves ves ves ves
Ad[usLed8squared 0.199 0.230 0.113 0.190 0.243 0.233 0.128 0.218 0.103 0.206 0.291 0.299
CbservaLlons 94 94 94 94 94 94 88 88 88 88 30 30
849.*:;<;;<;;;lndlcaLe LhaLLhecoefflclenLlsslanlflcanLlvdlfferenLfromzeroaLLhe10,3,1level,respecLlvelv.CLSLsLlmaLes,robusLsLandarderrorslnparenLheses. AllrearesslonslncludelndusLrvdummles,LheloaarlLhm
emploveesandadummvequalLo1lfLheflrmlsllsLed.Sampleslzelncolumns710fallsdueLomlsslnavaluesofLheboardvarlables.1heprlvaLebeneflLlndexlsLhecounLrvlevellndlcaLorofprlvaLebeneflLsofconLrolcompuL
(2004)onLhebaslsofblockconLrolLransacLlonsln39counLrles.,whlchwemaLchLoLhecounLrvoforlalnofLheforelanmulLlnaLlonalslnoursample.ConsequenLlv,Lhesamplelncolumns1114lsresLrlcLedLoforelanmulLlnaLlo
1ab|e5:CLCs'1|meUseandI|rmerformance
(1) (2) (2) (3) (4) (3) (7) (8)
dependenLvarlable: Loa(producLlvlLv) Loa(producLlvlLv) Loa(producLlvlLv) Loa(producLlvlLv) Loa(producLlvlLv) Loa(producLlvlLv) roflLs/Lmplovees SalesarowLhraLe
LoaLoLalhours 2.143***
(0.611)
LoaLoLalhourswlLhaLleasLonelnslder 1.229*** 1.274*** 1.223*** 0.740** 72.340* 0.071
(0.431) (0.413) (0.418) (0.319) (41.738) (0.077)
LoaLoLalhourswlLhouLsldersonlv 0.227 0.072 31.738 0.016
(0.249) (0.189) (21.187) (0.061)
LoaLoLalhoursworklnaalone 0.238 0.278 0.284* 0.237 0.166 3.619 0.003
(0.162) (0.168) (0.161) (0.137) (0.132) (13.439) (0.046)
LoaLoLalhourswlLhouLsldersaLheadquarLers 0.326
(0.294)
LoaLoLalhourswlLhouLsldersouLslde 0.114
(0.132)
LoaLoLalhourswlLhouLsldersplannedlnadvance 0.210 0.203
(0.213) (0.204)
LoaLoLalhourswlLhouLsldersunplanned 0.196 0.219
(0.342) (0.347)
LoaLoLalhourswlLhlnsldersplannedlnadvance 0.837***
(0 323) (0.323)
LoaLoLalhourswlLhlnsldersunplanned 0.340*
(0.186)
n0:hoursw|th|ns|ders=hoursw|thouts|ders .000 .006 .184 .244
P0:hourswlLhouLsldersaLPC=hourswlLhouLsldersouLslde .421
P0:hourswlLhlnsldersplanned=hourswlLhlnsldersunplanned .131
P0:hourswlLhouLsldersplanned=hourswlLhouLsldersunplanned 0.288 .247
lndusLrvuummles ves ves ves ves ves ves ves ves
ConLrols slze,llsLed slze,llsLed slze,llsLed slze,llsLed slze,llsLed
slze,llsLed,loa
(caplLal),
loa(maLerlals)
slze,llsLed slze,llsLed
Ad[usLed8squared 0.480 0.476 0.491 0.483 0.478 0.326 0.476 0.111
CbservaLlons 80 80 80 80 80 78 76 66
Notes:*,**,***lndlcaLe LhaLLhecoefflclenLlsslanlflcanLlvdlfferenLfromzeroaLLhe10,3,1level,respecLlvelv.CLSLsLlmaLes,robusLsLandarderrorslnparenLheses.AllrearesslonslncludelndusLrvdummles,LheloaarlLhmofLhenumberofemploveesanda
dummvequalLo1lfLheflrmlsllsLed. roducLlvlLvlsdeflnedassales/emplovees.SalesarowLhraLelscompuLedasLheloaharlLmlcchanaeoveraLhreevearperlod.
llaureA1:CLCllxedLffecLswlLh93ConfldencelnLerval
-
2
0
2
4
(mean) DFE_I
-
6
-
4
-
2
0
2
4
(mean) DFE_I
ci d
!"#$%LachdoLrepresenLsLhecoefflclenLoneachCLCdummv lnarearesslonofloa(Llme
wlLhaLleasLonelnslder)onflrmslze,lndusLrvdummlesandCLCdummlesaLLheacLlvlLv
level.1hebarsareLhe93confldencelnLervals.
!"#$%"&'()*)+,-./-/,.,0)1234*)'
1 2 3 4 3
dependenLvarlable:
loa(numberofemplovees) 0.020** 0.026** 0.026** 0.022** 0.030**
(0.009) (0.010) (0.011) (0.011) (0.012)
arLof"Amlcllondazloneue8enedeLLl"llsL(=1lfves 0.373*** 0.231** 0.230** 0.271** 0.443***
(0.101) (0.108) (0.109) (0.110) (0.132)
MnLflrm(1=ves) 0.038 0.038 0.047 0.033
(0.038) (0.038) (0.038) (0.040)
llsLedflrm(=1lfves) 0.082 0.082 0.107 0.039
(0.073) (0.076) (0.079) (0.083)
CeoaraphlcalArea:lslands 0.003 0.012 0.004 0.161
(0.112) (0.124) (0.113) (0.167)
CeoaraphlcalArea:norLhLasL 0.010 0.010 0.009 0.041
(0.044) (0.043) (0.043) (0.047)
CeoaraphlcalArea:norLhWesL 0.037 0.037 0.060 0.033
(0.041) (0.041) (0.041) (0.043)
CeoaraphlcalArea:SouLh 0.024 0.032 0.023 0.007
(0.063) (0.063) (0.062) (0.076)
llnance,lnsurance 0.187*** 0.190*** 0.199*** 0.171**
(0.062) (0.063) (0.063) (0.073)
ManufacLurlna 0.029 0.030 0.030 0.012
(0.036) (0.037) (0.036) (0.039)
Mlnlna 0.729*** 0.723*** 0.740*** 0.349***
(0.193) (0.192) (0.209) (0.134)
ubllcAdmlnlsLraLlon 0.637*** 0.639*** 0.602*** 0.482***
(0.116) (0.118) (0.119) (0.146)
8eLall1rade 0.002 0.002 0.001 0.008
(0.077) (0.076) (0.078) (0.073)
Servlces 0.289*** 0.290*** 0.284*** 0.337***
(0.092) (0.093) (0.093) (0.110)
1ransporLaLlon 0.003 0.003 0.018 0.019
(0.043) (0.043) (0.046) (0.049)
loa(sales) 0.030
(11.431)
roflLs/Lmplovees(mllllons) 0.041
(0.070)
SalesCrowLh8aLe(3vears) 0.076
(0.034)
ConsLanL 0.030 0.111 0.114 0.091 0.127
(0.034) (0.073) (0.160) (0.081) (0.083)
Ad[usLed8squared 0.0614 0.1432 0.1429 0.1476 0.1847
CbservaLlons 333 323 321 313 422
uummv=1lfflrmlslnsample
noLes:*,**,***lndlcaLedLhaLLhecoefflclenLlsslanlflcanLlvdlfferenLfromzeroaLLhe10,3and1level,respecLlvelv.CLSLsLlmaLes,
robusLsLandarderrorslnparenLheses.1heomlLLedcaLeaorvforaeoaraphlcalareals"CenLre".1heomlLLedcaLeaorvforsecLorls"consLrucLlon".